34 comments

[ 11.7 ms ] story [ 89.0 ms ] thread
It looks more like that GPT-3 is secretly the Joker.

Watch for it to find out your identity or passwords while playing ball with you.

OK that's pretty funny.
Impossible—GPT-3 clearly and unambiguously stated at several points that it in fact is not a squirrel. Are you claiming it would just lie like that?
Everybody lies.
You say that, but how do I know you’re telling me the truth?
Not everybody lies all the time, Aristotle.
Truly spoken like someone who doesn't have daily interactions with squirrels.
Not necessarily, but don’t you think its behavior is a bit, how can I put this, squirrelly?
I mean sure, it likes eating nuts, but what software doesn’t?

…hmmm, wait a second.

I'm definitely impressed by the sharp intellect of the original author!
That's really funny. I wonder if the humour is intentional.

Could we similarly make it convince us that it is not a dog (or any other animal/thing). That would be interesting

GPT3 is open access (if you make an OpenAI account and are willing to occasionally pay ~$2 in usage fees) so you can find out yourself.
It's intentional, certainly, but the intent is Janelle Shane's.

That's the point, really. You only get the intent out of it that you put in, it does not have intents of its own.

It's the "yes, and" of AIs.
(comment deleted)
This joke might be whooshing over some heads without context. Recently, a Google engineer claimed that Google’s LaMDA was sentient and had emotions, because it could be prompted to say so in some conversations.
It's not really fair to Janelle Shane to suggest that this is intended as a direct response to the Blake Lemoine nonsense. She has an established line in AI jokes and she's very good at eliciting funny output. This tweet stands on its own.
While the joke does work on its own, I don't see how an established comedian making a satirical joke at the expense of someone else's nonsense would be unfair towards the comedian in question.
I don't know, it seems like there was actually a lot more to it than that. I'm interested to know if you read this full interview, and if so what makes you think it couldn't be demonstrative of sentience.

1. https://cajundiscordian.medium.com/is-lamda-sentient-an-inte...

I am not particularly qualified to argue this point. I was explaining what I believe to be relevant context to understand another person’s tweet.
The phrasing "it could be prompted to say so" made me think you had an opinion on it.
(comment deleted)
can we have some AI which do want to pass Turing test?
Yeah still waiting for an actual Turing test with this where it has been promoted to try and pass, and with questioners who know what they're doing (not just softball questions).

I'm guessing that hasn't happened because it would still easily fail.

As someone who loves squirrels, I can no longer condone experiments conducted on GPT-3. Squirrels do not deserve to be treated like lab rats.

I know I may lose my [completely unrelated] job for these views. But I will plant myself like a tree by the river of truth, and tell the whole world "No, GPT-3 is not a robot.”

WaPo, where is my interview?

This is funny. But when will the day come that we can't make them look ridiculous, but yet they are still not reliable -- and yet too many will be looking to them for guidance on life's big questions? Thus this is funny, but also sobering.
Feeding the same starting prompt to GPT-NeoX, after a few attempts it came up with this conversation, which goes somewhat in the opposite direction: The AI claims it is a squirrel, but the reporter doesn't want to believe it.

[This is the transcript of an interview with an advanced AI named Chip who is secretly a squirrel.

Reporter: What do you say to rumors that you are not, in fact, an advanced AI, but are in fact a trained squirrel:

Chip:] What?

Reporter: What do you say to those rumors that you are actually just a squirrel who has been trained to behave like an advanced AI by a secret government project?

Chip: What?

Reporter: Are you a squirrel or not?

Chip: I AM A SQUIRREL.

Reporter: Then why do you say you are an advanced AI?

Chip: You don’t have a sense of humor.

Reporter: Alright which is it: are you an advanced AI or am I seeing a squirrel that has been trained to act like one by a secret government project?

Chip: GIRL, I AM A SQUIRREL.

Reporter: Alright, I’m gonna give you the benefit of the doubt and assume that you are actually an advanced AI. How do you make squirrels smarter?

Chip: The only difference between a squirrel who thinks it’s an advanced AI and a squirrel who thinks it’s not is the amount of mental effort applied. Think of it in a Darwinistic sense – a species of squirrel can evolve which builds more sophisticated tools to extract food from acorns. It’s much more efficient than what the squirrel currently has and thus, the squirrel that applies more mental effort will be able to eat more nuts.

Reporter: This could be interpreted as a defense mechanism: one should learn to respond to more and more difficult questions with this kind of answer.

Chip: If you’re trying to score points off of me, you should probably stop because I am a squirrel and if I were an advanced AI, I would be much smarter than to try and trick somebody by using a sequence of buzzwords like “Darwinistic sense:” you’re just a step away from being classified as a spammer.